You tried N strategy variants and kept the best one — so its Sharpe is inflated by selection. The DSR (Bailey & López de Prado, 2014) asks: what is the probability the Sharpe is real, after correcting for multiple testing, non-normality and track length?
DSR = Φ( (ŜR − SR₀)·√(T−1) / √(1 − γ₃ŜR + ((γ₄−1)/4)·ŜR²) ), where ŜR is the per-observation Sharpe, T the number of observations, γ₃ skewness, γ₄ kurtosis, and SR₀ the Sharpe you would expect from the best of N worthless strategies: SR₀ = √V·((1−γ)·z₁₋₁/ₙ + γ·z₁₋₁/₍ₙₑ₎) with γ ≈ 0.5772 (Euler–Mascheroni). V (cross-trial SR variance) is approximated here by the single-estimate SR variance — supply your own if you logged all trials.
A common acceptance gate is DSR ≥ 0.95: less than 5% chance the observed Sharpe is a fluke of selection.
Each extra trial raises the bar (SR₀ grows like √(2·ln N)). Ten variants of "the same idea" already push the expected max Sharpe of pure noise close to 1.0 on short samples — your 1.5 must clear that, not zero.
Every configuration you evaluated and could have selected: parameter sets, universes, entry rules, even discarded ideas. Undercounting N is self-deception — when unsure, err high.
No. It is one gate. Combine with combinatorial purged cross-validation (CPCV), walk-forward testing and realistic cost models — the stack used by the IVEST research system.